Semi-automatic quantification of the epicardial fat in CT images

In this work we present a technique to automatically or semi-automatically quantify the epicardial fat in noncontrasted Computed Tomography (CT) images. In CT images, the epicardial fat is very close to the pericardial fat, distincted only by the pericardium. The pericardium appears in the image as a very thin line, very hard to discriminate. To enhance the pericardium line and to remove noise as well as higher intensities due to calcifications, some pre-processing was applied, namely region growing, thresholding and average filtering techniques. To detect the pericardium line an algorithm was developed that considerer the heart anatomy to find control points belonging to that line. From the points detected an interpolation was done based on the cubic spline method. This method was also improved to avoid incorrect interpolation that occurs when one of the coordinates of the points is repeated. After having the line delineation, the pixels bellow the line were counted, considering only the pixels in the fat window (-190 to -30 Hounsfiel Units). In 10 images tested, in 4 the system fully automatically returned the correct value for epicardial fat. In the other 6 the system needed a small correction by moving 1 or 2 points to return the correct value of epicardial fat. The values of the automatic quantification were compared to the values obtained by the manual process, having 10% as maximum error allowed. We concluded that this method is able to, automatically or with a small interaction, return the value of the epicardial fat, for the non contrast CT images tested